stabilityai

/

sd-x2-latent-upscaler

like 169

I think this model have potential to achieve Midjourney or Imagen super resolution quality if with better finetune. But how?

I think this model have potential to achieve Midjourney or Imagen super resolution quality if with better finetune. But how?

I think the vae has a problem on recover face and hand details, maybe a better vae is needed? The diffusion model is well trained